TI 2005-113/2 Tinbergen Institute Discussion Paper Gigerenzer the Decided

Floris Heukelom

HME Department, University of Amsterdam, and Tinbergen Institute.

Tinbergen Institute The Tinbergen Institute is the institute for economic research of the Erasmus Universiteit Rotterdam, Universiteit van Amsterdam, and Vrije Universiteit Amsterdam.

Tinbergen Institute Amsterdam Roetersstraat 31 1018 WB Amsterdam The Netherlands Tel.: +31(0)20 551 3500 Fax: +31(0)20 551 3555

Tinbergen Institute Rotterdam Burg. Oudlaan 50 3062 PA Rotterdam The Netherlands Tel.: +31(0)10 408 8900 Fax: +31(0)10 408 9031

Please send questions and/or remarks of non- scientific nature to [email protected]. Most TI discussion papers can be downloaded at http://www.tinbergen.nl. Gigerenzer the Decided a tale of difficult distinctions

The journey ventures into a land of rationality that is different from the fam iliar one we know from m any stories in cognitive science and econom ics –tales in which hum ans live in a world with unlim ited tim e and knowledge, where the sun of enlightenm ent shines down in beam s of logic and probability. The new land of rationality we set out to explore is, in contrast, shrouded in a m ist of dim uncertainty. People in this world have only lim ited tim e, knowledge, and com putational capacities with which to m ake inferences about what happens in the enigm atic places in their world. - Gigerenzer , Todd and the ABC research group (1999)

I argue that the conceptual distinction between single-event probabilities and frequencies is of direct relevance for , and vice versa.

In short, the proper functioning of a m ental algorithm depends on the way in which inform ation is represented. So, to analyze probabilistic reasoning, we m ust attend to the difference between, at least, the frequency and the single-event representation of probability. If evolution has favored one of these form s of representations, then it would be frequencies, which prelinguistic organism s could observe and act on. Attending to this distinction suffices to m ake several apparently stable cognitive illusions disappear. - Gigerenzer (1994)

Gigerenzer the Decided – Floris Heukelom , August 2005

Introduction Hum ans are intriguing beings. They are also very com plex beings. The result of m illions of years of evolution, they have developed sophisticated m eans to deal with their environm ent. So where should we begin to investigate the working of these com plex beings? Fortunately, the com plexity partly lies therein that hum an beings have developed am azingly sim ple rules of thum b as a basis for their behavior. As a start one could study these fast and frugal heuristics. This, in a nutshell, is the view of the psychologist Gigerenzer. It is, of course, not an idea that cam e from nowhere, but is the result of years of research. Specifically, it is to a large degree shaped by his disagreem ent with another and closely related psychological theory of the working of the hum an m ind: the Heuristics and Biases theory of Kahnem an and Tversky. This disagreem ent itself was to a considerable extent shaped by Gigerenzer’s earlier work on the history of and , which sprang from his interests in psychophysical m odels of m easurem ent with which he started his career. As not everything can be treated in the detail it deserves, the focus in this paper will be on Gigerenzer’s criticism of Tversky and Kahnem an and his com peting research program as advanced since the late 1990s. After an introductory overview of the work of Gigerenzer in the first section, som e general them es of the background of Gigerenzer’s research are introduced in the second section. The third section provides an overview of Gigerenzer’s involvem ent in the collaborative research on the history of probability theory and statistics as it form s the basis of his criticism of Kahnem an and Tversky, treated in the fourth, and of his own research program , treated in the fifth section. After concluding rem arks have ended the paper, Tversky and Kahnem an’s answer to Gigerenzer’s allegations are sum m arized in an appendix.

1 Gigerenzer the Decided – Floris Heukelom , August 2005

1. Introductory overview of the work of Gigerenzer1 Born in 1947, Gigerenzer receives a degree in psychology from the University of Munich in 1974. Although 27 m ay seem relatively old to obtain a university degree, it is the average by Germ an standards. Three years later in 1977 Gigerenzer receives his PhD and five years after this, in 1982, his Habilitation, both from the University of Munich. In different positions Gigerenzer is affiliated with the University of Munich from 1974 up to 1984. After this he holds from 1984 until 1995, in chronological order, professorships at the universities of Konstanz (Germ any), Salzburg (Austria), and Chicago (USA). From 1995 to 1997 Gigerenzer is director of the Max Planck Institute for Psychological Research in Munich. Since 1997 he has been the director of the Max Planck institute for Hum an Developm ent in Berlin. The publications of Gigerenzer are characterized by a gradual shift from exclusively Germ an in the first four years to alm ost exclusively English in the first years of the twenty-first century. As a broad approxim ation Gigerenzer’s publications are furtherm ore characterized by an em phasis on m athem atical m easurem ent m odels in social psychology and psychophysics in the first five years, to an em phasis on (the history of) probability and statistics during the 1980s, to an em phasis on (bounded) rationality during the last fifteen years. Gigerenzer is m arried to Lorraine Daston, historian and director of the Max Planck Institute in the history of science2, with whom he professionally collaborated at the tim e of his work on the history of statistics during the 1980s.

2. Other introductory them es Gigerenzer is, at least in the early years of his career, both a psychophysician and a cognitive psychologist, a com bination he shares with for exam ple Kahnem an. The form alist or m athem atical character of the m easurem ent m odels of psychophysics dem ands an explicit

1 This introduction is drawn on the basis of Gigerenzer’s CV as it is available from the website of his research group, see http://www.m pib- berlin.m pg.de/en/forschung/abc/index.htm 2 Daston labels herself an historian of ideas, specializing in the m athem atics of the eighteenth century (Daston (2001), p.9).

2 Gigerenzer the Decided – Floris Heukelom , August 2005 opinion of the psychologist about the usefulness of m athem atics in psychological research. Before setting out the content of Gigerenzer’s work it seem s hence relevant to first introduce both psychophysics and cognitive psychology, and the view on m athem atics especially the form er im plies.

What is Psychophysics? Psychophysics, or physiological psychology as it is som etim es referred to3, is the scientific field that investigates the physiology of the senses. A total of seven senses is distinguished, but the larger part of psychophysical research concentrates nevertheless on the visual sense4. Psychophysics em ploys an experim ental m ethodology in which the relationship between physical stim uli and their responses in the form of sensational perception is investigated. Most of the tim e the sensation and the perception are equated, i.e. no perception m eans no sensation5. Som etim es, however, a distinction is drawn. In this case a sensation is m ore narrowly defined as the first stage in a chain of a (bio) chem ical and neurological process that m ay lead to a perception6. Psychophysics originates in the second half of the nineteenth century and counts am ong its m ain early figures such authors as Helm holtz, Fechner and Wundt7. As a result of his attem pt to construct scales for the perceived sensations and the resulting proposition of the just noticeable difference (jnd), especially Fechner stands out as a key figure. His psychological scales are taken to derive naturally from our everyday and experience. Hot and cold, for exam ple, are psychological scalings of tem perature stim uli. This basic fram ework of stim ulus-response and the notion of psychological scales have survived unchanged to this day. If we consider the psychophysical work of Gigerenzer, but also of Kahnem an and Tversky for instance, it is evident that the m ain question is still how to m easure the

3 See for exam ple Boring (1929). Physiological psychology in this sem inal history on experim ental psychology refers predom inantly to the pre-Fechnerian first half of the nineteenth century. 4 See the Wikipedia on psychophysics: http://en.wikipedia.org/wiki/Psychophysics 5 http://www.cogsci.princeton.edu/cgi-bin/webwn?stage=1&word=sensation 6 See the Wikipedia on sensation: http://en.wikipedia.org/wiki/Sensation 7 Murray (1993), Boring (1929), Kline (2000), Michel (1999), Danziger (1990), and m any others.

3 Gigerenzer the Decided – Floris Heukelom , August 2005 sensational responses given the stim ulus-response fram ework8. Gigerenzer him self describes psychophysics in slightly different term s as arising out of the divergence between our attem pt to describe the world around us, and the question of how we perceive that world:

“Psychophysik, so könnte m an sagen, entsteht aus der Divergenz zwischen der physikalischen Beschreibung der Welt und der Beschreibung derselben m ittels unserer Wahrnem ung.”9

The stim ulus-response fram ework of psychophysics and its related discussion on m easurem ent has been taken over by behaviorism and cognitive psychology in the twentieth century. As a result, it is difficult to distinguish on a m ethodological basis between the psychological research program s of psychophysics, (neo)- behaviorism , m athem atical psychology, and cognitive psychology. This is not to say there has not been any disagreem ent about the right m ethod for the stim ulus-response m easurem ent. What it does im ply is that it is problem atic to fram e these different views on the m easurem ent of stim uli and responses in term s of the well-known branches of psychology. Put differently, it is often im possible to put on the basis of the m ethodology em ployed a distinctive label of a psychological sub-field on publications by psychologists as Kahnem an, Tversky and Gigerenzer. In an historical account that focuses on the m ethodologies em ployed these term s therefore necessarily rem ain som ewhat vague10.

What is cognitive science? It is notoriously difficult to pin down exactly what entails cognitive science. A sensible first step could be to look up the definition of cognition as provided by the Oxford Dictionary. Cognition is:

8 One m ight object that in recent years the stim ulus-response fram ework is increasingly replaced with a gam e theoretic approach. I would fully agree. 9 Bredenkam p and Gigerenzer (1984) 10 This is a m inor point here, but m ay surface m ore clearly would one at a later stage want to pinpoint m ore precisely the difference between the two cam ps as described in this paper in m ethodological term s.

4 Gigerenzer the Decided – Floris Heukelom , August 2005

“The action or faculty of knowing taken in its widest sense, including sensation, perception, conception, etc., as distinguished from feeling and volition; also, m ore specifically, the action of cognising an object in perception proper.”11

Taking the definition literally hence im plies that cognitive science m ust be the scientific investigation of knowing, taken in its widest sense but distinguished from feeling and volition. And indeed, as a first approxim ation this seem s a reasonable definition. In a next step this definition can be com pared with definitions of cognitive science as provided by authoritative sources. Two of the m ost used of these sources are Wikipdia, the free and biggest online encyclopedia, and the Stanford Encyclopedia of Philosophy. This yields the following:

“Cognitive science is usually defined as the scientific study either of m ind or of intelligence.”12 and,

“Cognitive science is the interdisciplinary study of m ind and intelligence, em bracing philosophy, artificial intelligence, neuroscience, linguistics, and anthropology.”13

There is thus a subtle difference between the use of the of cognition in everyday English, and cognition as it is understood by scientists in the collocation cognitive science. The scientific use points in the direction of a science that investigates intelligent beings in their totality, the everyday definition of cognition points in the direction of cognitive science investigating knowledge, or rational behavior, as opposed to non-rational behavior such as em otions. However, if we read a bit further into the description of cognitive science as provided by the Stanford Encyclopedia , we find that “[t]he central hypothesis of

11 http://dictionary.oed.com 12 See the Wikipedia on cognitive science: http://en.wikipedia.org/wiki/Cogntive_Science 13 Thagard (2004)

5 Gigerenzer the Decided – Floris Heukelom , August 2005 cognitive science is that thinking is best understood in term s of representational structures in the m ind and com putational procedures that operate on those structures.”14 (m y em phasis), which suggests that em otions, feelings and so on, do not form part of cognitive science. It m ust hence be that either cognitive science considers em otions, feelings and the like to be non-existing relics of the past, or that it considers these to be included in the study of the m ind and of intelligence. At first sight this m ay seem to be a m inor issue. The reason that I stress it is that I believe that it m ay very well be at the basis of m uch of contem porary debate in cognitive science and cognitive psychology, including the controversy between Gigerenzer, and Kahnem an and Tversky. The basic question, then, is the following: if one investigates the working of the m ind of intelligent beings, does that only include that part of behavior that can be deem ed ‘rational’, or does it include also the behavior which is the result of em otions, passions, drifts, feelings, volition, and so on? We shall com e back to this question below. Let us first further specify the cognitive sciences. A sub branch of cognitive science is cognitive psychology. With respect to cognitive science generally it restricts itself to the study of (intelligent) behavior of hum an beings, as opposed to for exam ple the investigation of intelligent behavior of chim panzees, dolphins and robots. It furtherm ore restricts itself to the individual, thereby ignoring theories of super-hum an phenom ena such as cultures. That said, the relation with other sciences such as artificial intelligence and econom ics is im m ediate. The interdisciplinary em phasis of cognitive science also is a characteristic of cognitive psychology. As m entioned, the m ethodology em ployed by cognitive science and cognitive psychology is the psychophysical experim ent of stim ulus and response. Subjects in laboratories are given (physical) stim uli after which the corresponding response is m easured by the experim enter or reported by the subject. In contem porary cognitive psychology these are often called judgm ents of the stim uli. The biggest part of the cognitive psychological literature hence consists of theories of how the stim ulus given to the subjects relates to the responses m easured or judgm ents m ade.

14 Thagard (2004)

6 Gigerenzer the Decided – Floris Heukelom , August 2005

This brings us to the description of what cognitive science/psychology is not, which is perhaps the clearest exposition of what it entails. Cognitive psychology theorizes about the hum an m ind in its totality. In its totality, it views the m ind as a process that gives a certain kind of response when given a certain stim ulus. What it does not investigate is how different parts of the brain relate to the behavior of the individual. The work of Dam asio (c.f. 1994, 2003) is thus typically not part of cognitive science. What cognitive psychology neither does is investigate how m ore than one individual m ay act together in a group, or what the nature and functioning is of the institution in which we trade the goods we produce. Cognitive science, in other words, is neither sociology or econom ics, nor neuroscience. That said, it should be added the different fields m ay be com bined, which is precisely what we see happening at this m om ent. Cognitive psychologists m ay use neuroscientific knowledge as a basis, a constraint, a source of inform ation for their cognitive psychological theories. Econom ists and sociologists m ay use descriptions of hum an beings from cognitive psychology for their theories about hum an interaction and m arkets. Finally, som e com m ents should be m ade about the m etaphors em ployed by cognitive science and cognitive psychology. The m ost com m on m etaphor in cognitive science is that of the hum an m ind as a com puter. For instance, m an’s cognitive capacities are often called com putational capacities, and theories about the functioning of the hum an m ind are often stated in the typical if/then language of com puter’s software. This does however not im ply that the m ind is equated with a com puter15, if only because there is no such thing as the com puter. Another m etaphor that has risen to som e prom inence in cognitive psychology is the m etaphor of the m ind as an intuitive statistician. Metaphors and their upgrading to full theories of m ind in cognitive science and cognitive psychology are an im portant elem ent in the work of Gigerenzer. Cognitive scientists in general, however, seem to be aware of the m etaphors they use.

15 Although som e cognitive scientists com e pretty close. The m ost visible discussion on cognitive science’s use of the com puter as a m etaphor of the m ind is the debate following Searle’s criticism on what he labelled ‘strong AI’ in his well-known ‘Chinese room ’ article. See Searle (1980), Boden (1988), and others.

7 Gigerenzer the Decided – Floris Heukelom , August 2005

Gigerenzer on m athem atics During the first few years of his academ ic career an im portant part of Gigerenzer’s publications deal with m athem atical m ethods of m easurem ent in psychophysics16. This research is in the sam e vein as, or is at least closely related to, the theoretical work on m easurem ent by Tversky. In Gigerenzer (1977) for instance he m entions the paper by Tversky and Krantz (1969) on the experim ental additivity of interdim ensional characteristics as an exam ple of the research he him self is also involved in. The view that com es out of these early publications is one in which m athem atics in psychology generally speaking is worth pursuing as it forces the scientist to be precise and consistent. Although the em phasis on m athem atical m odels of m easurem ent, and psychophysics generally, disappears in later years, it does not seem that Gigerenzer’s ideas about m athem atics have changed. That is, in Gigerenzer’s continuous view psychological theories, classifications, m easurem ent m odels and so forth, should be put in m athem atical term s if possible17. As an exam ple of his view of m athem atics, consider the next quote from an early paper on taxonom y and taxonom etry (taxonom y by m eans of m athem atics).

“m it m athem atischen Methoden [können] die Kapazitätsschwäche unserer Sprache und dam it unserer Urteilsschem ata ausgeglichen und die Kom m unikation über Verhaltensstrukturen zuverlässig verbessert werden.”18

3. Gigerenzer and the history of probability and statistics In 1975 Ian Hacking published his The Em ergence of Probability. This book, which discusses the first few decades in which the m athem atical interpretation of the concept of probability was taken at hand by such scholars as Pacal, Ferm at and the Bernoulli’s, soon becam e a classic.

16 See for exam ple Gigerenzer (1977a, 199b, 1984,1986), Bredenkam p and Gigerenzer (1984) and Gigerenzer and Sarris (1982) 17 Gigerenzer’s recent attem pt to get into contact with econom ics could perhaps partly be explained by the wish to becom e once again m ore form al in his theories. 18 Gigerenzer (1977a), p.740

8 Gigerenzer the Decided – Floris Heukelom , August 2005

Not only did it becam e the m ain point of reference for those writing on the history of probability and statistics, it was also an im portant catalyst for further research. In the years following the publication of this book, the historical investigation of probability theory and statistics was taken up by a num ber of researchers who over the years frequently collaborated in a range of publications, sem inars, and conferences. As a very rough and im precise approxim ation, the m ain contributors to this collaborative research can considered to be Lorenz Krüger, Michael Heidelberger, Lorraine Daston, Mary Morgan, David Murray, Ted Porter, Kurt Danziger, John Beatty, and . The first in this series of publications following Hacking is Probability and conceptual change in scientific thought (1982), a book edited by Heidelberger and Krüger and com prised of different contributions by researchers who participated in a workshop in Bielefeld in July 1981. The Em pire of Chance (1989), a collaboration of Gigerenzer, Swijtink, Porter, Daston, Beatty, and Krüger which puts together all expertise on the subject, acts as the pinnacle of the collaborative research, after which the different contributors went their separate ways. Only a few years after finishing his PhD, Gigerenzer thus becam e involved in a broad collaborative interest in the history of probability theory and statistics. In the context of this collaboration som e specific expertise of Gigerenzer in this research can be pointed out and som e personal opinions or em phasis can be unraveled. It is nevertheless difficult to tell exactly where Gigerenzer starts and the collaborative research ends (and vice versa). Although Gigerenzer stands out with respect to the other contributors in that he rem ains throughout prim arily a practicing experim ental psychologist, it is problem atic, if not right out im possible to tell whether a certain view or description of probability reflects a general shared view of probability, or a personal view of Gigerenzer on how to use probability in psychology. Furtherm ore, the literature on the history of probability and statistics generally is perm eated with a strong sense of what constitutes good science and good psychology. Often it is not clear whether this is a reflection of writing on this particular subject, or of an attem pt to use historical argum ents to advance a specific view on how to do science/psychology. Put differently, given the fundam ental character

9 Gigerenzer the Decided – Floris Heukelom , August 2005 of probability theory and statistics and the fact that it has penetrated each and every corner of our life (including our language), the idea that even an attem pt at historical objectivity is doom ed to fail should be kept in m ind. Gigerenzer, in other words, is statistics and probability in psychology, and Gigerenzer is history of probability theory and statistics. Despite all these shortcom ings there is m uch to tell. The best way to proceed, then, seem s to be to give an account of the general them es in the historical writings on probability and statistics, which then function at the sam e tim e as general them es in the work of Gigerenzer. Given a selection of the of historical accounts of probability theory and statistics19, these them es are, to be sure, what I infer to be the m ain issues. In this account I will consider the history of probability and statistics generally known.

Probability, statistics and rational m en The rise of what becam e known as classical probability in the period between roughly 1650 and 1830, is a them e in the larger story of the Enlightenm ent20. The classical probabilists set them selves to the task of describing in m athem atical term s both the beliefs of the newly conceived rational m an and the frequencies of worldly events this rational enlightened m an wanted to explain. From the first m om ent onwards the m athem atical interpretation of probability therewith had a dual m eaning. The interpretation of a duality in the m eaning of probability is due to Hacking (1975), and taken over by other authors writing on the subject21. Following Hacking it is often referred to as the Janus-faced

19 i.e. Hacking (1975), Heidelberger , Krüger (1982), Heidelberger, Krüger and Rheinwald (1983) Daston (1982, 1986), Danziger (1990,1997), Krüger, Gigerenzer and Morgan (1987), Krüger, Daston and Hedelberger (1987), Gigerenzer and Murray (1987), Gigerenzer, Swijtink, Porter, Daston, Beatty and Krüger (1989), Porter (1986) Stigler (1986), Klein and Morgan (2001), Kurz-Milcke and Gigerenzer (2004). 20 Here I do not elaborate on the relation between science, m athem atics and the Enlightenm ent. See for careful expositions on these subjects for exam ple Outram (1995), Daston (1986) and Ham pson (1968). 21 The first to m ake this dual interpretation, however, was Carnap (1944). Som e, like Daston (1986) distinguish between m ore than two m eanings of probability, an issue that unfortunately I have to leave to be treated elsewhere. See for exam ple Gigerenzer (1994)

10 Gigerenzer the Decided – Floris Heukelom , August 2005 character of probability. On the one hand, probability in the seventeenth century referred to a rational degree of belief. It was, to put it in Hacking’s term s, an “epistem ological” notion “dedicated to assessing reasonable degrees of belief”22. The dom ain of this notion was prim arily the courtroom in which an ever-recurring problem was (and is) when to convict an individual if certain evidence is lacking but it is nevertheless ‘probable’ that the accused has com m itted the crim e. On the other hand, probability referred to the attem pt to produce stable, long run frequencies. With the rise of insurances, annuities, and large scale lotteries there was an increasing interest in, and need for, constructing reliable m easurem ents of death rates, ship returning rates, and so forth. Also these frequencies were put in term s, and understood to be, probabilities. The Janus-faced interpretation of the m eaning of probability by Hacking and others does of course not im ply that it was at the tim e understood as such. On the contrary, the beliefs of rational (or reasonable) m en should correspond to the calculated frequencies of real world events. If they did not som ething had to be wrong with the calculation of the frequencies. During the tim e of classical probability the m athem atical description of the beliefs of rational m an becam e increasingly problem atic. The well known exam ple here is the St. Petersburg paradox which intrigued scholars for m any years. The reason for this is that it so clearly showed the discrepancy between the optim al action according to the m athem atical logic em ployed, and the action a rational m an was going to take. It was intuitively clear to everyone that no rational m an would bet m ore than a seventeenth century’s equivalent of four or five euros on a coin flipping gam ble with a 2n euro pay-off. This im plied for the m athem atical probabilists a problem with the m athem atics, not with the rational beliefs of educated m en. The proposed solution of Daniel Bernoulli to distinguish between m athem atical and m oral expectation, was hence an attem pt to save the m athem atics. But there was a second reason why the attem pt to describe the intuitive beliefs of rational m an in m athem atical term s failed, which can best be seen in the light of a general critique on the concept of the enlightened rational m an. As the discussion in classical probability

22 Hacking (1975), p.12

11 Gigerenzer the Decided – Floris Heukelom , August 2005 advanced throughout the eighteenth century it was increasingly the notion of rationality itself that was questioned. It becam e less and less clear what exactly entailed rationality, and which m en could be said to be rational with respect to their beliefs. In first instance, rationality referred to the intuitive beliefs of educated (enlightened) m en. But after it was shown that not only did educated m en held different beliefs, but also that educated m en far from always acted according to their rational beliefs, rationality becam e interpreted as referring to the beliefs of all hum an beings23. The interpretation of rationality shifted to the idea that all beliefs by hum an beings were fundam entally rational but that through religion, superstition, education and so forth these rational beliefs were corrupted. When also this interpretation becam e problem atic the concept of rational intuitions of the enlightened m an were abandoned as incorrect. From this m om ent on m athem atics concentrated on the second m eaning of probability (that of the description of frequencies of real world events). The breakdown of the classical probability of the Enlightenm ent thus coincides with the rise of statistics. The disentanglem ent of the frequencies of real world events from the intuitions of rational m an enabled this m eaning of probability a rapid rise to prom inence. The reason is that the m athem aticians in their description of frequencies of real world events no longer were bound by the idea that these frequencies should som ehow correspond to rational beliefs. Instead they could freely gather data and discuss the best way to construct frequencies. Thus the nineteenth century becam e the century of the rise of statistics. This rise of statistics is characterized by what Stigler labels a vertical and a horizontal developm ent24. Vertical in the sense that the m athem atics of statistics itself greatly advanced and in the sense that this m eaning of probability becam e m uch better understood, horizontal in the sense that statistics spread to an increasing num ber of sciences by which it changed in fundam ental ways. After a century of developm ent of the m athem atics of statistics, it was Kolm ogorov who in 1933 provided an axiom atic basis for the m athem atics of statistics. From that m om ent on the m athem atics of

23 Ham pson (1968), p.195 24 Stigler (1986), p.4

12 Gigerenzer the Decided – Floris Heukelom , August 2005 statistics was largely agreed upon. The question of how and where to apply statistics, however, was far from settled. Especially in econom ics and psychology, where statistics becam e only of serious influence during the first half of the twentieth century, the debate fiercely raged. As in the other sciences in which statistics had risen to dom inance, the debate in psychology in first instance circled around the question whether the nature of the subject of scientific inquiry was determ inistic or indeterm inistic25. As for instance Maxwell had argued in the case of physics, a m inority of psychologists began to argue the fundam ental indeterm inistic character of the psychological science. The two m ost prom inent scholars advocating indeterm inism in psychology were Thurstone and Brunswik, and it is these two authors that are the historical speciality of Gigerenzer. Let us consider them in som e m ore detail.

Determ inism vs. indeterm inism in psychology The work of the psychologists Thurstone and Brunswik is analysed by Gigerenzer in the light of the discussion between determ inism and indeterm inism in psychology, and western science generally. To be sure, probability theory and statistics are not necessarily connected only with indeterm inism . On the contrary, until well into the nineteenth century probability and statistics neatly fitted with the determ inistic worldview held. The rationale behind em ploying probability and statistics in a determ inistic world is that although an event can be predicted with certainty if all causes and relevant factors are known, in reality this is hardly ever possible. Thus, in a determ inistic worldview it is held that if one knows all the relevant factors influencing the coin tossed, there is no surprise as to which side it is going to land on. In reality, however, we cannot with enough precision, if at all, determ ine all the relevant influencing factors and their values. Probability theory and statistics are here useful because they provide predictions of frequencies and probabilities in the absence of full inform ation of the causes of an event.

25 Although closely related, the rise of statistics in econom ics is different of character from that in psychology. For histories of statistics in econom ics see for exam ple Klein and Morgan (2001),… ..

13 Gigerenzer the Decided – Floris Heukelom , August 2005

In an indeterm inistic worldview, on the other hand, it is held that probability is a fundam ental character of our world. It is argued that som e processes are of an irreducible probabilistic nature. This im plies that knowing all the causes does not provide us with certainty about the event. For a num ber of reasons the idea of an indeterm inistic world m et with a lot of opposition. For one, as the business of science had always been to produce certain knowledge in a world of superstition and false beliefs, it was one thing to say that our knowledge of the world would always be im perfect. It was quite another thing to say that som e phenom ena were driven by chance and that nothing definite could ever be said about their causes. Maxwell personifies the indeterm inistic worldview for the natural sciences. In psychology Thurstone and Brunswik were the m ain advocates of an indeterm inistic approach. Gigerenzer m akes a distinction between two branches in the (Am erican) psychology of the first half of the twentieth century: on the one hand the experim ental psychology which em ployed the stim ulus- response fram ework which originated in nineteenth century Germ an psychophysics, on the other hand the correlational psychology that used correlational statistics after K. Pearson26. The m ain area of interest of experim ental psychology was perception, of correlational psychology it was intelligence. On a few grounds these two psychologies differed fundam entally from one another. Experim ental psychology studied the individual and tried to construct objective m ethods for m easuring the sensational responses given by the subjects in the laboratory. It held a fully determ inistic view of psychology that was created and m aintained in the im age of classical physics. The experim ents were set up in a laboratory and used the (physics’) laboratory m ethod of isolation and control. Thus, m any potential stim uli were controlled, after which one stim ulus was varied and its corresponding sensational response m easured. Correlational statistics on the other hand, used the correlational statistics of K. Pearson to construct the psychological characteristics of an average m an. It would go to far, however, to label this an indeterm inistic

26 This division corresponds with Danziger’s (1990) experim ental vs. applied psychology distinction. To be sure, the experim ental vs. correlational division was the standard division m ade in early twentieth century (Am erican) psychology.

14 Gigerenzer the Decided – Floris Heukelom , August 2005 approach to psychology27. The need to use correlational statistics was considered to result from the inherent im perfectness of our knowledge of nature. Nevertheless, the use of statistics in constructing averages was central to its m ethod and to its view on how to do psychology. The m ain difference between correlational and experim ental psychology is not so m uch a difference between determ inism and indeterm inism , but a difference of the subject of investigation. Experim ental psychology investigated the individual in the laboratory, correlational psychology the psychological characteristics of an average m an. From 1927 onwards, both Brunswik and Thurstone advocated a way to reconcile the two opposing cam ps, Brunswik unsuccessfully, Thurstone successfully28. The alternative contained two m ain elem ents. Firstly, with respect to experim ental psychology it was proposed to take the subject out of the laboratory and study it in its environm ent. Brunswik and Thurstone therewith attem pted to com bine the focus on the individual and the use of stim ulus-response of the experim entalists with the conviction of the correlationalists that the psychological subjects should be investigated in their natural environm ent without any constraints. Secondly, Brunswik and Thurstone proposed to use statistics to solve the problem of the instable environm ent of these stim ulus-response experim ents in the real world. More specifically, they proposed to consider the stim ulus perceived by the individual as a draw from a distribution (norm al as a first approxim ation). The assum ption was that the presentation of an object (say a light bulb or a sound) does not always lead to the exact sam e psychological value of the stim ulus. When an individual is presented with, say, two sounds of which he has to decide which is the louder, the probability he will say a to be louder than b will depend on the overlap of the two distributions of the stim uli. In this respect it com bined the correlationalists conviction that psychology should concern itself with averages with the experim entalists focus on

27 Gigerenzer, Swijtink, Porter, Daston, Beatty and Krüger (1989),p. 60 28 Gigerenzer is m ainly interested in Brunswik. Although the work of the two, of course, differs, I attem pt to sum m arize in the following the m ain propositions (on which they agreed, or so it seem s) without going in the precise details of the argum ents m ade.

15 Gigerenzer the Decided – Floris Heukelom , August 2005 individuals. The averaging of the responses given by the subjects functioned as a solution for the assum ption of the distributional character of the subject’s responses. It is in this respect that Brunswik spoke of subjects as “intuitive statisticians”29. In the proposed m ethod of m easuring responses in psychology by Thurstone and Brunswik, the hum an subject was the central figure. The experim ents all depended on how, and with what m agnitude the subject perceived the stim ulus. Although the object was a given, the responses of the individuals differed because there are not two individuals exactly the sam e and differences will occur as to the m agnitude of the stim ulation resulting from the object presented. “Intuitive statistician” in this regard is not a statem ent about the constitution of the individual’s m ind but a statem ent of how to think of individual responses with respect to an aggregation of individual responses30. The psychological m easurem ent m ethod advanced by Brunswik and Thurstone did not survive and gave way to the m easurem ent theory as form ulated by S. Stevens. Stevens did away with the intuitive statistical part of the m easurem ent and sim ply stated the m agnitude of the stim ulus to be equal to the m easured m agnitude of the object presented. If the volum e of a tone was 50 dB, the m agnitude of the stim ulus was 50 dB. In a way he thus denied the individuality of the subjects. He did, for instance, not consider the possibility that som e individuals had better ears than others. Subjects becam e interchangeable neutral m easurem ent instrum ents. It also im plied that the subjective processes that m ade the stim ulus result in a certain response could be considered independently of the m ethods and instrum ents of m easurem ent. The em phasis therewith shifted to the (m athem atical) analysis of the m easurem ent data. The concept of the “intuitive statistician”, invented by Brunswik as a reference to the distributional character of the perceived stim ulus, disappeared with the rest of the theory but m ade an unexpected re-appearance in the

29 Despite the fact that Gigerenzer puts the discussion on Brunswik and Thurstone repeatedly in a determ inism versus indeterm inism story, it rem ains som ewhat vague (to m e at least) whether these authors should be considered indeterm inists. In Gigerenzer (1987a, 1987b) his answer seem s to be yes. In Gigerenzer (1994) it is decidedly no. 30 Or so I conclude on the basis of Gigerenzer texts.

16 Gigerenzer the Decided – Floris Heukelom , August 2005 years following the Second World War when Am erican psychologists gave a new interpretation to the term in their analyses of judgm ent under uncertainty.

The re-em ergence of rational m an The period of the 1930s to the 1960s that saw the rise of statistics in psychology and econom ics also saw an as yet unexplained31 re- em ergence of rational m an. Again was posed the idea of a rational m an that would take decisions according to the optim al solution as calculated by m athem atics. The notion of probability used this tim e to determ ine the optim al decision was the subjectivist (or Bayesian, or epistem ic) interpretation of probability, m ixed with the utility m axim izing interpretation of rationality as advanced by von Neum ann and Morgentern32. Rationality thus was given a rather specific m eaning and deviations from the m athem atically determ ined optim al rational behavior becam e quickly interpreted as (irrational) errors. But for whatever reason the re-em ergence of rational m an occurred, it was (again) a wrong turn. Its narrow focus on one type of probability com bined with a useless understanding of rationality form s the basis for Gigerenzer’s critique on Kahnem an and Tversky. The dom inant notion of rationality used by so m any social scientists also functions as the benchm ark by which Gigerenzer contrasts his own cognitive psychological theory. In the following we will first address Gigerenzer’s critique on Kahnem an and Tversky, after which we will turn to Gigerenzer and his rationalities.

4. Gigerenzer’s critique on Kahnem an and Tversky Gigerenzer’s critique of the cognitive psychology of Kahnem an and Tversky runs along two closely related lines: statistics and rationality. His critique is both an extension of his historical work on statistics, as a negative definition of the psychological research program m e he later starts to advance. Gigerenzer’s work can hence both them atically and

31 That is, unexplained according to Gigerenzer c.s.. In view of for exam ple Carnap (1944), logical positivism could be a first place to look. 32 I would suggest that also elem ents of the Keynesian/Ram seyan theory of rational partial beliefs play a role. This is however not accounted for in the expositions of Gigerenzer c.s..

17 Gigerenzer the Decided – Floris Heukelom , August 2005 chronologically be organised as: history of statistics => critique on Kahnem an and Tversky => program m e. The intim ate connection of these three them es in the work of Gigerenzer is reflected in his publications, which hardly ever deal with just one of his interests. The publications used for this section33 can nevertheless be considered as the core of his critique on Tversky and Kahnem an. It should furtherm ore be noted that not all psychologists consider the difference as fundam ental as Gigerenzer. Generally speaking cognitive psychologists seem to consider the difference between Gigerenzer and Kahnem an and Tversky to be a difference of degree, and not so m uch a fundam ental difference of insights34. One of the things Gigerenzer cannot stress enough is that the theory of probability or statistics does not exist. Most scientists are fam iliar with the difference between the Bayesian, or subjectivist, and the frequentist, or objectivist, school in statistics. These two schools in statistics often arrive at different solutions to statistical problem s. But it runs deeper than that. The fundam ental difference of opinion of R. Fisher, and Neym an and E. Pearson on how to do som ething as fundam ental as hypothesis testing, for instance, has never been solved. Alm ost all textbooks on statistics after the Second World War have ignored this difference and have presented statistics as one consistent and unattested body of knowledge, a fallacious account of past and present debate in statistics. The fact that Kahnem an and Tversky only use Bayesian statistics is hence a narrow focus that is constraining and unnecessary. The fact that they use Bayesian statistics to construct the optim al (or norm ative, or rational) decision really m isses the point, according to Gigerenzer. If highly educated m athem aticians do not agree on the best way to apply statistics, it is absurd to suppose that individuals should behave according to one of these m ethods. But even within the realm of the Bayesian statistics Tversky and Kahnem an m ake assertions that are at best only part of the truth. In this regard the recurring exam ple in Gigerenzer’s work is that of the

33 i.e. Gigerenzer (1991,1993,1994,1996), Gigerenzer, Hell and Blank (1988), Gigerenzer and Murray (1987), Sedlm eier and Gigerenzer (1997,2000), Hertwig and Gigerenzer (1999) 34 Frey and Benz (2004)

18 Gigerenzer the Decided – Floris Heukelom , August 2005 phenom enon of base rate neglect, as discovered by Tversky and Kahnem an. In different experim ents Kahnem an and Tversky show that people often neglect the base rate distribution of a population they have to use to solve a problem . A well-known experim ent is the cab- accident problem , which runs as follows. A witness has seen how in the m iddle of the night a cab caused an accident and drove away before anyone could register its num ber plate. In the town there are only blue (25%) cabs and green cabs (75%). The witness is positive he has seen a green cab, but experim ents under sim ilar conditions have shown he is only right in 80% percent of the cases. What is the probability the cab indeed was green? Kahnem an and Tversky show that individuals fail to take the base rate of the cabs (25% blue, 75% green) into account and thus com m it the fallacy of base rate neglect. Gigerenzer, however, argues that the answer depends on which base rates one takes. Should one take the percentages of each colour cab in town? Or should one for instance use the base rate of the percentages of each colour cab involved in accidents? Or the percentages of each colour that work at night? Or in the specific part of town in which the accident occurred? One can, in other words, not accuse the individual of ignoring one pair of base rates when it is not clear which base rates one should use in the particular case. To use another illustration of this point by Gigerenzer, if you are invited to the dean’s party and are asked to guess the profession of a to you unknown 50 year old m ale, it does not seem a good strategy to base your guess upon the country’s base rates of m ale professions. A-select draws in reality alm ost never occur. The critique on Kahnem an and Tversky’s can be placed under the heading of Gigerenzer’s critique on biases generally. In a range of publications Gigerenzer’s shows that the cognitive illusions of Tversky and Kahnem an can be m ade to “disappear, reappear, or even invert”35. He em ploys both theoretical and experim ental argum ents to m ake his case. The base rate neglect as m entioned above functions in Gigerenzer’s work as a good exam ple of how a bias can be shown to disappear when the theory upon which it is built is carefully scrutinized. But also for instance the overconfidence bias and the conjunction bias can be m ade to

35 Gigerenzer (2000), p.243

19 Gigerenzer the Decided – Floris Heukelom , August 2005 disappear. In both these cases Gigerenzer shows that the reported bias is a result of the experim ental setting. In the case of the overconfidence bias it is shown that individuals distinguish between single events and long run frequencies. If subjects are given questions of the type: which city has m ore inhabitants a) Hyderabad, or b) Islam abad, and have to indicate how confident they are their answer is correct (choosing between 50,60,70,80,90 and 100%), their average confidence is indeed higher than the average of correct answers. However, if they are asked how well they did after an experim ent with fifty such questions, their estim ates are, on average, practically the sam e as the true average of correct answers. Thus, Gigerenzer argues that hum an beings show no overconfidence bias when faced with frequencies. This only leaves a supposed overconfidence bias in the case of single events, which is dism issed by Gigerenzer on theoretical grounds. Building on the authority of Richard von Mises and de Finetti, Gigerenzer rem inds the reader that probability applied to a single event is m eaningless. The argum ents against, and falsifications of, the follow a sim ilar pattern. If subjects are asked for frequencies they do pretty well, (subjective) probabilities of single events m ake no sense. Placing Gigerenzer’s critique on the Biases of Kahnem an and Tversky under an again m ore general heading it is to be understood as resulting from Gigerenzer’s disagreem ent with what he labels ‘the norm ative issue’. In the norm ative-descriptive fram ework as em ployed by Kahnem an and Tversky, observed judgm ents are com pared with a norm . But what is norm ative here? Gigerenzer shows that on the basis of probability theory and statistics it is not possible to construct one correct answer (or one norm ) in judgm ents under uncertainty. There is furtherm ore far from any agreem ent on which situations probability and/or statistical inference m ay be applied to. A unique norm , in other words, cannot be constructed and a norm ative decision theory is hence m eaningless36.

36 In the Nobel prize interview (Kahnem an (2002)) Kahnem an m ore or less concedes on this point by noting that the reason he and Tversky put so m uch em phasis on the norm ative issue was the hope and belief in the 1970s of ultim ately being able to construct an unam biguous tool for (political/adm inistrative) decision m aking.

20 Gigerenzer the Decided – Floris Heukelom , August 2005

The norm ative decision theory which is the result of conceiving of the hum an m ind as a (Bayesian) intuitive statistician has for som e tim e37 been used by Gigerenzer as an exam ple of a philosophy of science theory labelled tools-to-theories. The idea is that what initially starts out as a tool (som etim es) becom es a theory of its own. In case of Edwards, Kahnem an, Tversky and others the tool of the m etaphor of the intuitive statistician as it was originally proposed by Brunswik transform s into a theory of the m ind as an intuitive statistician. The ‘norm ative issue’, then, naturally follows from this transform ation. Needless to say Gigerenzer disagrees with tools transform ing into theories. But Gigerenzer’s attack after attack on the norm ative issue is also the result of a m ore general dissatisfaction with the project of cognitive psychology, or so I would argue. What lies at the background of Gigerenzer’s criticism of Tversky and Kahnem an c.s. is the focus on decisions that are m ade on the basis of cognition. Gigerenzer does not agree with cognitive psychology leaving out other characteristics of hum an psychology, like em otions, passions, and so forth. What lies at the bottom of Gigerenzer’s ideas of how to do cognitive psychology is the conviction that when hum an decision m aking is investigated, the entire hum an being should be considered, not just a part of it. This provides the clue to the final, and to m y m ind m ost fundam ental, criticism on Kahnem an and Tversky, nam ely on their use of the concept of rationality; as well as to the research program on the different rationalities as Gigerenzer starts to prom ote it from roughly 1997 onwards. When Kahnem an and Tversky start their collaborative research there already exists quite an extensive literature on how hum an beings deal with probability and statistics38. However, all this research follows Savage (1954) in that it only investigates ‘sm all world’ situations. A ‘sm all world’ in Savage’s theory is a controlled laboratory setting in which exists clear agreem ent on the probabilities used. The ‘sm all world’ theory only considers problem s for which the cognitive capacities of the m ind are required. Just like the cognitive sciences generally it does not argue that all behavior is cognitive or rational,

37 Roughly on and off since 1987 38 For an overview see Peterson and Beach (1967)

21 Gigerenzer the Decided – Floris Heukelom , August 2005 but sim ply restricts itself to the behavior in which no em otions and so forth are involved. Exam ples include the judgm ents which of two light bulbs is brighter, or the judgm ent of which am ount of m oney is m ore preferred. Kahnem an and Tversky apply this theoretical fram ework to real- life situations. The Heuristics and Biases theory they propose is an extrapolation of intuitive statistics research to the real world. Kahnem an and Tversky im plicitly argue that every situation in life that involves in one way or another probabilistic and/or statistical inference can be analysed in term s of the theory put forth by Savage39. Therewith, Gigerenzer argues40, Tversky and Kahnem an reason that all decisions hum an beings m ake in real life that involve uncertainty are m ade on the basis of cognition, or rationality. By im plication deviations from the calculated optim al decision thereby becom e irrational errors or fallacies. This then is at the basis of Tversky and Kahnem an’s theory, and equally at the basis of Gigerenzer’s critique. The em ployed definition of rationality Kahnem an and Tversky take over from Savage’s sm all world theory assum es subjects to correctly calculate, given all the relevant inform ation, the optim al solution that m axim izes pay-off, utility or whatever. Apart from the fact that in real life this would often take far too long, it also ignores em otions, passions etc.. Furtherm ore it leads to behavior that we never observe. In this respect, the relevant passage in Cognition as Intuitive Statistics (1987), finally, is worth quoting at length:

“In sum , the belief that Bayes’ theorem or another theory of statistical inference is rational thinking em pties rationality of its m ost interesting aspects. Rationality is without purpose, without weights for conflicting goals, without inform ation search and exploration, without judgm ents of relevance, without reflection on content, without use of contextual inform ation and cost- benefit considerations. To the disappointm ent of philosophers like Popper, it also leaves out the context of discovery, the truly

39 Indeed, one of the m ain criticism s in the decision theory cam p on Kahnem an and Tversky is precisely this un-authorized extrapolation of the ‘sm all world’ theory to the real world. 40 That is to say, in m y reading of Gigerenzer.

22 Gigerenzer the Decided – Floris Heukelom , August 2005

psychological problem of creativity explored by the Würzburg and Gestalt schools. And the fascination with such m echanical rationality tends to divorce the psychology of inductive thinking from all these aspects, if only by dism issing them as sources of irrationality. Our view is that the task of a psychology of thinking is to understand how these various sources direct hum an thought rather than to judge whether thinking is rational or not.” (p.181)

5. Gigerenzer and his rationalities The dissatisfaction of Gigerenzer with the cognitive psychology of Kahnem an and Tversky is the result of a general dissatisfaction with cognitive psychology. His objection is principally the fact that cognitive psychology is only concerned with the cognitive part of psychology, instead of the hum an being in its entirety. Gigerenzer does not disagree with the idea that som e part of our behavior can be labelled ‘cognitive’, but with the fact that such a narrow focus ignores the influence of other psychological characteristics. When in 1997 Gigerenzer becom es the director of a Max Planck research group in Berlin he com es into a position in which he can advance his own approach. Roughly from this m om ent on his dissatisfaction with the standard paradigm , as well as his proposed alternative, is form ulated in term s of rationality. His critique on the dom inant interpretation of the individual calculating the optim al response or action and behaving accordingly is that it is unrealistic. In reality people do not have all the relevant inform ation to calculate the optim al action, nor the capacities to do so (especially if we consider the lim ited tim e in which judgm ents and decisions often have to be m ade). The em phasis on the cognitive part of hum an behavior has led cognitive psychologists to theories in which every judgm ent is m ade and every decision taken as if it were deliberate, cognitive, rational behavior. This is a m istaken view according to Gigerenzer. Gigerenzer form ulates his alternative in term s of other, different, types of rationalities. In this, he explicitly follows the bounded rationality fram ework as set out by Sim on. In the bounded rationality account of hum an behavior as it is put forth by Sim on, behavior of hum an beings should be understood as the result of two

23 Gigerenzer the Decided – Floris Heukelom , August 2005 phenom ena. These two phenom ena are understood to cut together like a pair of scissors. They are necessary com plem ents. On the one hand we find the incom pleteness of the inform ation upon which hum ans have to base their judgm ent and decisions, as it is in reality never the case that we posses all the relevant inform ation for the decision to be taken. The other blade of the pair of scissors consists of the lim ited com putational capacities of hum an beings. We sim ply do not have the capability to put all the available inform ation together in a careful and com plex calculation. Hum an beings are rational in the sense that they try to m ake the best judgm ent or decision. They are however only boundedly so given both the incom pleteness of the inform ation and the lim itness of their com putational capacities41. Then how does the decision and judgm ent m aking happen given these two blades of scissors? Gigerenzer argues, again in the spirit of Sim on, that this happens on the basis of heuristics. A crucial factor that Gigerenzer adds is that these heuristics are fast and frugal. The idea is that in our decision and judgm ent m aking we hum ans em ploy heuristics that lead to quick response and dem and as little com putational capacities as possible. In contrast to the heuristics as advanced by Kahnem an and Tversky, Gigerenzer’s heuristics are not sim plified versions of com plex (rational) decision m aking rules but sim ple rules of thum b that work well in a certain context. To use the sam e exam ple as before, in an often used experim ent of Gigerenzer subjects are asked for a series of pairs of cities which is the larger one in term s of inhabitants. He shows that subjects use a list of cues on which they base their decision. The first cue is whether the subject knows both cities. If only one city is known, or far better known, the process stops and that city is decided to be the larger one. If both cities are equally well known, a second cue m ay for exam ple be if the soccer clubs of both cities are known. If the soccer club of one city is (far better) known than the soccer club of the other city, than that city is taken to be the larger one. And so forth. Individuals thus use fast and frugal heuristics that yield the best result for a given decision or judgm ent problem .

41 In line with Gigerenzer’s earlier historical work as well as in line with his critique on Kahnem an and Tversky, the bounded rationality he favours is explicitly defined as not involving utilities and probabilities.

24 Gigerenzer the Decided – Floris Heukelom , August 2005

Gigerenzer’s research at his centre for Adaptive Behavior and Cognition (ABC) investigates how and why these heuristics (som etim es described as form ing part of an ‘adaptive toolbox’) function. The how question investigates which heuristics are used in which situations and how a choice is m ade between different heuristics. The exam ple described above of the heuristic involved in choosing the larger of two cities is an exam ple of this research. But Gigerenzer is also interested in answering why these heuristics function so well. In an extension of the two-city experim ent he conducts for instance an em pirical investigation of the correlation between the size of a city and how often it is m entioned in a newspaper. As these correlations are fairly high he thus shows why the sim ple rule of thum b works so well. Given an understanding of hum ans and their environm ent in term s of bounded rationality, Gigerenzer investigates the heuristics hum ans em ploy to deal with the incom plete inform ation and lim ited com putational capacities at their disposal. Because such a definition is still too broad, in his research institute he has m ade a sub-division of four areas of research (along a fifth called ‘Methods, Metaphors, and Theory Construction’ which elaborates on his tools-to-theories research), nam ely 1) bounded rationality, 2) , 3) social rationality, and 4) evolutionary rationality. The bounded rationality research area explores the general structure as set out above. The other three rationalities each explore different structures that determ ine the type or pattern of inform ation the subjects face. Ecological and evolutionary rationality reflect the idea that over the course of our evolution our heuristics have evolved in such a way as to optim ally em ploy the structure of our environm ent. Research in the evolutionary branch (which is situated in, or linked with, ) investigates how and why these heuristics evolved over (evolutionary) tim e. The research conducted under the heading of ecological rationality investigates these heuristics them selves. Social rationality, on the other hand, reflects the idea that hum ans also need heuristics to deal with the quickly changing structure of our social environm ents. Research under this heading investigates these heuristics. At the m om ent of writing it is the focus on these last three types of rationalities, which em phasize the type of the environm ent in which

25 Gigerenzer the Decided – Floris Heukelom , August 2005 the decision or judgm ent has to be m ade, that m ost effort is put into. It thus seem s as if the em phasis on the heuristics them selves has shifted som ewhat to the investigation of the role of the environm ent in the decision and judgm ent m aking process. Another indication in that direction is Gigerenzer’s recent interest in linking (his) psychology with other sciences, notably behavioral biology42 and econom ics43. It seem s (but this is speculation) that Gigerenzer has com e to the conclusion that all sciences that have som ething to say about behavior should be linked together. A less friendly interpretation is that he wants to advance his program in as m any areas as possible. What seem s to happen in any case at this m om ent is that Gigerenzer is offering his psychological worldview to econom ist as a better alternative to base econom ic theories upon, and is looking to biology to sharpen his hum an decision m aking theories and to provide them with a biological grounding.

Conclusion This paper has provided an overview of the work of the psychologist Gigerenzer, in which the focus has been on the m ore visible parts of his research. Gigerenzer starts his academ ic research in the late 1970s as a cognitive psychologist whose m ain interest is on questions of m easurem ent in psychophysics. In the early 1980s he becom es involved in a broad collaborative research program on the history of probability theory and statistics, which lasts from roughly 1981 to 1989. Gigerenzer’s criticism of the psychology of Kahnem an and Tversky is to a large extent shaped by this historical research. At the risk of exaggeration one could sum m arize Gigerenzer’s criticism s in the verdict that Kahnem an and Tversky do not understand probability, statistics and rationality. Gigerenzer’s own research program at the ABC in Berlin from 1997 onwards continues in turn on his critique of Tversky and Kahnem an. It is an im plication of it. Gigerenzer starts where he argues Tversky and Kahnem an go wrong. His alternative builds on the bounded rationality work of Sim on in which decision-m aking should be understood as the interaction of im perfect inform ation and lim ited

42 e.g. Hutchinson and Gigerenzer (2005) 43 e.g. Gigerenzer and Selten (2001)

26 Gigerenzer the Decided – Floris Heukelom , August 2005 com putational capacities. In recent years Gigerenzer attem pts to surpass the fram ework as set out by Sim on by distinguishing between different types of rationalities and by linking his research program with other sciences that m ake claim s about hum an behavior. The psychologies of Gigerenzer on the one, and of Kahnem an and Tversky on the other hand are both virtually the sam e and fundam entally different. Both approaches consider decision m aking the m ain approach to investigate hum an psychology. Both also agree that hum an decision m aking is directed by heuristics and offer explanations on how and why these heuristics work, and how they relate to one another. However, it is im possible to theorize about decision m aking without utilities and probabilities in Kahnem an and Tversky’s worldview, where in Gigerenzer’s worldview it is im possible to theorize about decision m aking with utilities and probabilities. Despite Gigerenzer’s convincing and decided rhetoric, the dispute rem ains as yet undecided.

27 Gigerenzer the Decided – Floris Heukelom , August 2005

Appendix: Tversky and Kahnem an’s response to Gigerenzer’s allegations The highest trees catch the m ost wind is the Dutch saying. It is thus not surprising that the theory advanced by such visible cognitive psychologists as Kahnem an and Tversky has been attacked, disproved and adjusted by a wide range of authors. However, Tversky and Kahnem an have only responded to what has probably been their m ost vigorous opponent, and even this response am ounted to no m ore than one article and a very brief postscript. It is nevertheless difficult to sum m arize what the response in On the Reality of Cognitive Illusions (1996) exactly entails. Also Gigerenzer’s im m ediate reply On Narrow Norm s and Vague Heuristics: A Reply to Kahnem an and Tversky (1996) (1996) does not add a lot to the clarification of the dispute. To a considerable extent this has to do with the fact that both parties m ore or less evade a discussion. The m ain accusation of Kahnem an and Tversky is their being m isrepresented by Gigerenzer. In their fierce rhetoric this is put in the following term s:

“Gigerenzer’s reports on out work and on the evidence cannot be taken at face value. [..] The position described by Gigerenzer is indeed easy to refute, but it bears little resem blance to ours. It is useful to rem em ber that the refutation of a caricature can be no m ore that a caricature of a refutation.” (p.584)

Gigerenzer answers by arguing that the real problem does not lie on the level of the current (theoretical, em pirical) discussion, but on a philosophical level.

“I welcom e Kahnem an and Tversky’s (1996) reply to m y critique [..] and hope this exchange will encourage a rethinking of research strategies. I em phasize research strategies, rather than specific em pirical results or even explanations of those results, because I believe that this debate is fundam entally about what constitutes a good question and a satisfactory answer in psychological research on reasoning.” (p.593)

28 Gigerenzer the Decided – Floris Heukelom , August 2005

The discussion that does develop consists of two points. Firstly Kahnem an and Tversky retort Gigerenzer’s argum ent that the biases disappear when the questions are form ulated in term s of frequencies by pointing out that along Bayesian probabilities for singular events they have also studied judgm ent of frequencies. They refer to a list of publications in which they show that also in judgm ents of frequencies people com m it m istakes. In a second argum ent under the heading “The Norm ative Issue”, Kahnem an and Tversky argue that in contrast to what Gigerenzer repeatedly shows, ‘bias’, ‘fallacies’, or ‘m istakes’ are real and thus indicate the existence of som e norm .

“This position [of Gigerenzer –FH], which m ay be characterized as norm ative agnosticism , is unreasonable perm issive. Is it not a m istake for a speaker to assign probabilities of .99 both to an event and to its com plem ent? We think that such judgm ents should be treated as m istaken; they violate accepted constraints on the use of probability in everyday discourse.” (p.586)

As said, the m ain part of Gigerenzer’s reply consists of arguing that the real difference between him and Kahnem an and Tversky consists of a difference in opinion of which philosophy of science practice to follow. His response to Kahnem an and Tversky’s accusation of him m isrepresenting them is that Kahnem an and Tversky in turn m isrepresent and fail to understand him . The final and m ost fundam ental criticism of Gigerenzer on Kahnem an and Tversky then is that the latter’s heuristics and biases are post hoc explanations that prove everything and nothing, and cannot be falsified. Indeed, in his reply Gigerenzer in the end com m its him self to a Popperian falsificationist, point of view. In the last sentence of the last postscript he thus states:

“As I see it, there are two ways in which a theory can fail: by being wrong or by being indeterm inate. The latter m ay be worse for scientific progress, because indeterm inate theories resist attem pts to prove, disprove, or even im prove them . Twenty-five years ago, extending on Ward Edward’s work, Kahnem an and

29 Gigerenzer the Decided – Floris Heukelom , August 2005

Tversky opened up a fertile field. Now it is tim e to plant theories.” (p.596)

30 Gigerenzer the Decided – Floris Heukelom , August 2005

Bibliography Boden, M. A. (1988). Escaping from the Chinese Room . Com puter Models of Mind. M. A. Boden. Cam bridge, Cam bridge University Press: Ch 8.

Boring, E. C. (1929). A History of Experim ental Psychology. New York, The Century Co.

Bredenkam p, J. and G. Gigerenzer (1984). "Einführung: Einige Gedanken zur Kontextabhängigkeit der Wahrnehm ung und des Urteils." Psychologische Beiträge 26: 91-101.

Carnap, R. (1944). "The Two of Probability." Philosophy and Phenom enological Research: 513-532.

Danziger, K. (1990). Constructing the Subject, historical origins of psychological research. New York, Cam bridge University Press.

Danziger, K. (1997). Nam ing the Mind, how Psychology found its language. London, SAGE Publications.

Daston, L. (1983). Rational Individuals versus Laws of Society. Probability since 1800. M. Heidelberger, L. Krüger and R. Rheinwald. Bielefeld, Universität Bielefeld: 7-26.

Daston, L. (1986). Classical Probability in the Enlightenm ent. Princeton, Princeton University Press.

Daston, L. and K. Park (2001). Wonders and the order of Nature, 1150- 1750. New York, Zone Books.

Frey, B. S. and M. Benz (2004). From im perialism to inspiration: a survey of econom ics and psychology. Elgar com panion to econom ics and philosophy. A. Marciano, J. Davis and J. Runde. Cheltenham , Edward Elgar: 61-83.

31 Gigerenzer the Decided – Floris Heukelom , August 2005

Gigerenzer, G. (1977a). Mathem atische Methoden zur Klassifikation von Personen. Binet und die Folgen. Die Psychologie des XX. Jahrhunderts. G. Strübe. Zürich, Kindler. Band V: 738-759.

Gigerenzer, G. (1977b). Nichtm etrische Dim ensionsanalyse. Binet unde die Folgen. Die Psychologie des 20. Jahrhunderts. G. Strübe. Zürich, Kindler. Band V: 713-737.

Gigerenzer, G. (1982). On the Role of Probability in Psychology: L.L. Thurstone's Solution to the Problem of Measurem ent and its Im pact on Psychological Research Today. Probability and Conceptual Change in Scientific Thought. M. Heidelberger and L. Krüger. Bielefeld, Universität Bielefeld.

Gigerenzer, G. (1984). "Lässt sich die Flächenwahrnehm ung als "kognitieve Algebra" beschreiben?" Psychologische Beiträge 26(113- 119).

Gigerenzer, G. (1986). "Wissenschaftliche Erkenntnis und die Funktion der Inferenzstatistik. Anm erkungen zu E. Leiser." Zeitschrift für Sozialpsychologie 17: 183-189.

Gigerenzer, G. (1987a). The Probabilistic Revolution in Psychology - an Overview. The Probabilistic Revolution 2. L. Krüger, G. Gigerenzer and M. S. Morgan. Cam bridge, MIT Press: 7-10.

Gigerenzer, G. (1987b). Probabilistic Thinking and the Fight against Subjectivity. The Probabilistic Revolution 2. L. Krüger, G. Gigerenzer and M. S. Morgan. Cam bridge, MIT Press: 11-33.

Gigerenzer, G. (1987c). Survival of the Fittest Probabilist: Brunswik, Thurstone, and the Two Disciplines of Psychology. The Probabilistic Revolution 2. L. Krüger, G. Gigerenzer and M. S. Morgan. Cam bridge, MIT Press: 49-72.

32 Gigerenzer the Decided – Floris Heukelom , August 2005

Gigerenzer, G. (1991). "How to Make Cognitive Illusions Disappear: Beyond "Heuristics and Biases"." European Review of Social Psychology 2: 83-115.

Gigerenzer, G. (1993). The bounded rationality of probabilistic m ental m odels. Rationality, Psychological and Philosophical Perspectives. K. I. Manktelow and P. E. Over. London, Routledge: 284-313.

Gigerenzer, G. (1996). "On Narrow Nom rs and Vague Heuristics: A Reply to Kahnem an and Tversky (1996)." Psychological Review 103(3): 592-596.

Gigerenzer, G. (2000). Adaptive thinking: rationality in the real world. New York, Oxford University Press.

Gigerenzer, G. and D. J. Murray (1987). Cognition as Intuitive Statistics. London, Lawrence Erlbaum Associates.

Gigerenzer, G. and V. Sarris (1982). "Psychophysik heute: Aktuelle Problem e und Ergebnisse." Psychologische Beiträge 24: 313-351.

Gigerenzer, G. and R. Selten (2001). Bounded rationality: the adaptive toolbox. Cam bridge, Cam bridge University Press.

Gigerenzer, G., Z. Swijtink, et al. (1989). The Em pire of Chance, Cam bridge University Press.

Gigerenzer, G. and P. M. Todd (1999). Fast and Frugal Heuristics, The Adaptive Toolbox. Sim ple Heuristics That Make Us Sm art. G. Gigerenzer, P. M. Todd and A. R. Group. New York, Oxford University Press: 3-34.

Gigerenzer, G., P. M. Todd, et al. (1999). Sim ple Heuristics That Make Us Sm art. Oxford, Oxford University Press.

Hacking, I. (1975). The Em ergence of Probability. London, Cam bridge University Press.

33 Gigerenzer the Decided – Floris Heukelom , August 2005

Ham pson, N. (1968). The Enlightenm ent, An evaluation of its assum ptions, attitudes and values. London, Penguin Books.

Heidelberger, M. and L. Krüger, Eds. (1982). Probability and Conceptual Change in Scientific Thought. Bielefeld, Universität Bielefeld.

Heidelberger, M., L. Kruger, et al., Eds. (1983). Probability since 1800, Universität Bielefeld.

Hertwig, R. and G. Gigerenzer (1999). "The 'Conjunction Fallacy' Revisited: How Intelligent Inferences Look Like Reasoning Errors." Journal of Behavioral Decision Making 12: 275-305.

Hutchinson, J. M. C. and G. Gigerenzer (2005). "Sim ple Heuristics and rules of thum b: Where psychologists and behavioural biologists m ight m eet." Behavioural Processes 69: 97-124.

Kahnem an, D. (2002). Daniel Kahnem an – Interview. Stockholm , Royal Swedish Academ y of Sciences.

Kahnem an, D. and A. Tversky (1996). "On the Reality of Cognitive Illusions." Psychological Review 103(3): 582-591.

Klein, J. L. and M. S. Morgan, Eds. (2001). The Age of Econom ic Measurem ent. Durham and London, Duke University Press.

Kline, P. (1998). The New Psychonom etrics. London, Routledge.

Krüger, L., L. J. Daston, et al., Eds. (1987). The Probabilistic Revolution Volum e 1: Ideas in History. Cam bridge, MIT Press.

Krüger, L., G. Gigerenzer, et al., Eds. (1987). The Probabilistic Revolution Volum e 2: Ideas in Science. Cam bridge, MIT Press.

34 Gigerenzer the Decided – Floris Heukelom , August 2005

Kurz-Milcke, E. and G. Gigerenzer, Eds. (2004). Experts in Science and Society. Dordrecht, Kluwer Academ ic/Plenum Publishers.

Michell, J. (1999). Measurem ent in Psychology. Cam bridge, Cam bridge University Press.

Murray, D. J. (1993). "A perspective for viewing the history of psychophysics." Behavioral and Brain Sciences 16: 115-186.

Outram , D. (1995). The Enlightenm ent. Cam bridge, Cam bridge University Press.

Peterson, C. R. and L. R. Beach (1967). "Man as an Intuitive Statistician." Psychological Bulletin 68(1): 29-46.

Porter, T. M. (1986). The Rise of Statistical Thinking 1820-1900. Princeton, Princeton University Press.

Searle, J. R. (1980). "Minds, Brains, and Program s." The Behavioral and Brain Sciences 3: 417-424.

Sedlm eier, P. and G. Gigerenzer (2000). "Was Bernoulli Wrong? On Intuitions about Sam ple Size." Journal of Behavioral Decision Making 13: 133-139.

Sedlm eier, P. and G. Gigerenzer (1997). "Intuitions About Sam ple Size: The Em pirical Law of Large Num bers." Journal of Behavioral Decision Making 10: 33-51.

Stigler, S. M. (1986). The History of Statistics, The Measurem ent of Uncertainty before 1900. Cam bridge, Harvard University Press.

Thagard, P. (2004). Cognitive Science. Stanford Encyclopedia of Philosophy. E. N. Zalta.

35 Gigerenzer the Decided – Floris Heukelom , August 2005

Tversky, A. and D. H. Krantz (1969). "Sim ilarity of schem atic faces: A test of interdim ensional additivity." Perception & Psychophysics 5(2): 124-128.

Wikipedia (2005). Cognitive Science, Wikipedia www.wikipedia.org.

Wikipedia (2005). Psychophysics, Wikipedia www.wikipedia.org.

Wikipedia (2005). Sensation, Wikipedia www.wikipedia.org.

36